knowledge: chess-probe sweep complete — vector-tenant ladder (96-bit > cam_pq confirmed), episodic keys GREEN, escalation GREEN+fence, rung honest [H]

[AdaWorldAPI]

Summary

Records the final four probes of the perturbation-cascade synthesis (stockfish-rs companion PR carries the code). Doc + board only. Three GREEN, one honest INCONCLUSIVE; the design space from operator input is recorded with honest grades.

Grades recorded

• D-SF-V3-4 (escalation) → GREEN (structure) + FENCE: king move ⟺ coarsest-tier change ⟺ full refresh, exact (4851/4851 vs 0/70,229); the literal RouteAction::Escalate name-match is a rhyme (that variant means percentile ambiguity) and is fenced.
• D-PALETTE-NNUE-VEC (tenant-shape ladder) → GREEN at the Base17 vector-compression anchor on held-out columns. The measured ladder — cam_pq 48-bit 0.929 < V3-L4 96-bit 6×(256×256) 0.966 < 64×256 "one centroid per square" 0.977 < +4bit turbovec-floor residue 0.998 (near-orth 0.994) — confirms the operator ruling that the full 96-bit pair-tenant beats 48-bit cam_pq. Anchor correction banked (0.999 Fisher-z = distance codec, wrong reference class for vector compression). Memorization-degeneracy lesson banked (train/eval split mandatory).
• D-SF-ARIGRAPH-1 (episodic keys) → GREEN (representation): 3.7× chance, 96% of raw cosine retained in the store's own [u64;8]/Hamming shape — plus the named GAP: EpisodicMemory today is string-only FNV label_fp with no fingerprint-in retrieve_similar; the probe's SimHash constructor is the missing primitive (queued).
• D-SF-RUNG-1 (hindsight) → INCONCLUSIVE, stays [H]: the NNUE-eval-only oracle is not mate-aware (sacrificial mates invert the label) and random playouts never settle — a measured negative on the oracle, honestly recorded.

Design space recorded (operator input)

• classid × wide-FieldMask in-memory filtering — zero-serialization, per-focus reasoning masks over resident rows (ClassView-projection doctrine); queued design note.
• holograph 7×7×7 signed bits — information-theoretically roomy, consumer-UNPROVEN → [H].
• Pythagorean-comma NON-collapsing Morton-cascade pyramid — the anti-mush mechanism (D-MTS-5 comma witnesses), 4096 → 256k×256k upstream replayability without materialization; the 3DGS 10000×10000 ~440 ms top-k figure recorded as operator-recalled precedent, [S]-here (perturbation-sim splat.rs/morton2 substrate is shipped; the figure is not reproduced in this repo's source).

Changes

• .claude/knowledge/stockfish-nnue-as-perturbation-cascade.md — the four deliverables graded with results; the ladder table incl. the 64×256 lane; the design-space paragraph; the two temporal-table rows updated.
• .claude/board/EPIPHANIES.md — E-CHESS-PROBE-SWEEP-COMPLETE-1 prepended (same commit), with the substrate-thesis roll-up: GREEN = temporal stream, palette-distance, Morton-gridlake, episodic basins, escalation structure, vector-tenant ladder, episodic-key representation; FENCED/[H] = RouteAction literal, rung-hindsight oracle, holograph, scalar-256 (superseded). The doc remains SYNTHESIS-not-canon; every grade carries its probe.

Testing

Doc + board only; the measurements are the stockfish-rs probes (companion PR), all fmt/clippy clean and CI-safe.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Awg6TXocHcwTtc6eGsHcdD

---

Generated by Claude Code

Summary by CodeRabbit

• Documentation
  • Added results from four parallel chess-engine probes, including escalation behavior, vector similarity, episodic retrieval, and NNUE evaluation.
  • Documented measured vector “tenant-shape” comparisons and identified a preferred 96-bit configuration over a 48-bit alternative.
  • Clarified that escalation is driven by ambiguity and distance thresholds rather than tier changes alone.
  • Recorded limitations and follow-up gaps for episodic-memory locality and hindsight evaluation, including oracle failures and inconclusive results.

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_8125eb37-4db3-4de1-aae9-8bc8688397cc)

[coderabbitai]

📝 Walkthrough

Walkthrough

Documents four measured probes covering escalation, NNUE vector tenants, episodic retrieval, and hindsight. It also clarifies fenced claims, records oracle failures, identifies an episodic locality gap, and adds a dated board roll-up.

Changes

Probe findings and substrate synthesis

Layer / File(s)	Summary
Palette and escalation findings .claude/knowledge/stockfish-nnue-as-perturbation-cascade.md	Adds measured NNUE tenant-shape comparisons and clarifies that escalation ambiguity triggers are distinct from the proven king-move/full-refresh structure.
Temporal and episodic probe findings .claude/knowledge/stockfish-nnue-as-perturbation-cascade.md	Records hindsight oracle failures, SimHash episodic-key results, and the missing locality-preserving retrieval interface.
Measured sweep board entry .claude/board/EPIPHANIES.md	Adds the dated four-probe summary, findings, status roll-up, and probe references.

Estimated code review effort: 2 (Simple) | ~15 minutes

Possibly related PRs

• AdaWorldAPI/lance-graph#679: Introduced related Stockfish NNUE perturbation-cascade and temporal/episodic substrate documentation.

Poem

A bunny hops through probes of green,
With measured paths and gaps unseen.
NNUE shapes align in flight,
SimHashes cluster bits just right.
Though hindsight’s oracle lost its way,
The board records the findings today.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately summarizes the main documentation update and the four probe outcomes.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In @.claude/knowledge/stockfish-nnue-as-perturbation-cascade.md:
- Around line 123-144: Add a blank line immediately before and after the
tenant-SHAPE ladder table in the D-PALETTE-NNUE-VEC section, keeping the table
content unchanged so Markdown parses it independently from the surrounding
prose.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 03a74f26-484d-418e-9e9f-fc43530bdfbf

📥 Commits

Reviewing files that changed from the base of the PR and between 5cc881e and efef496.

📒 Files selected for processing (2)

• .claude/board/EPIPHANIES.md
• .claude/knowledge/stockfish-nnue-as-perturbation-cascade.md